openJiuwen proposes MANGO for multi-agent flow networks

解锁Agent Swarm新潜力，openJiuwen又一力作：多智能体流网络

openJiuwen proposed MANGO, a multi-agent flow-network framework that combines reinforcement learning, textual gradients, and a Skip-k mechanism; using GPT-4o-mini, it reports a 12.8% accuracy gain over MaAS on MATH500 and a 5.1% F1 gain over AFlow on DROP.

Why it matters: HKR-K is strong: the post gives mechanisms and a MATH500 delta. HKR-H/R pass for the multi-agent flow-network angle, but this remains a research-framework story, not a major model or platform release.